Figure 2. Definition and evaluation workflow of regulated on-off-slide models.

On-off-slide models consist of processes from different hierarchies: (A) Global processes for assembly, disassembly and (optional) sliding. Global processes govern all reactions of the corresponding type with the same rate $r_A$, $r_D$ or $r_S$. To fit multiple promoter states simultaneously, some processes have to be regulated, that is, have different rate values depending on the promoter state. In this example, the global assembly process is regulated. (B) Optional site-specific processes for assembly and disassembly at each position (example here with rates $r_{A3}$ and $r_{D1}$) and for sliding between each neighboring pair of positions (here $r_{S12}$). Reactions in gray have not been overruled by more specific processes (here: site-specific processes) and consequently are still determined by the rate parameters of processes on the less specific hierarchy level (here: global processes). (C) The last hierarchy level is given by optional configuration-specific processes governing only one reaction (here with rates $r_{D1-4}$ and $r_{S4-3}$). Here, only the promoter configurations 1 to 4 are shown. (D) Each regulated on-off-slide model is fitted and evaluated successively using the experimental data on the left-hand side (promoter states during the experiment given in parentheses). Models are discarded if they do not match the maximum likelihood threshold after each stage. With each additional experimental data set, the fit results in new optimal relative rate values of the model. Only the dynamic Flag-/Myc-tagged histone measurements enable us to also fit the time scale.

Figure 2—figure supplement 1. Occurrences of the different processes in the models with satisfactory likelihood at the different stages.

(A) In all 68,145 analyzed models, (B) in all 173 good models of stage 1, (C) in all 15 good models of stage 2, (D) in all seven satisfactory models after stage 3. In each plot, the y-axis limit is the number of the considered models allowing the comparison of the relative occurrences between the four cases.

Figure 2—figure supplement 2. Stage 1: logarithmic likelihood ratio histograms.

Histograms of the logarithmic likelihood ratio with respect to the perfect fit likelihood in stage 1 (${\displaystyle R_1}$), that is, using the configurational data of Brown et al., 2013 only. (A-D) For all models and three model subsets. 0 on the x-axis corresponds to a perfect fit. Dashed green line: value of the best model (within the subset). Red line: threshold for a satisfactory fit of $R_{max}=6$. Sliding processes are not needed to find agreement with the measured configuration statistics alone. Two models with lower complexity, that is, with one fitted parameter less, are below the threshold $R_{max}$.

Figure 2—figure supplement 3. Stage 1: top 30 models with likelihood above the threshold.

The top 30 models with likelihood above the threshold after the first stage. The colored boxes in each row show the model processes and their rate values. White boxes denote the absence of a process in a model. Regulated processes are separated into two differently colored boxes for repressed (left half) and activated (right half) promoter state. Weakly activated rate values are not shown here. On the left side are the model number and the log10 ratio of the best possible likelihood and the model likelihood, ${\displaystyle R_1}$.

Figure 2—figure supplement 4. Stage 1: models with likelihood above the threshold and only six fitted parameters.

Same as in Figure 2—figure supplement 3, but showing satisfactory stage 1 regulated on-off-slide models with complexity reduced by one, that is, models with up to only six fitted parameters.

Figure 2—figure supplement 5. Reactions involving the sticky N-3 position.

Reactions involving the N-3 position. 'a3’, 'd3’, 's23’, and 's32’ denote sets of reactions. For each model the reaction rates are governed by the model’s processes as in stage 1. Simultaneously, to test models for the agreement with the sticky N-3 mutation experiments, the model’s reaction rates of assembly at N-3 (a3), disassembly at N-3 (d3), sliding from N-2 to N-3 (s23) and sliding from N-3 to N-2 (s32), each obtain a prefactor whose values are found by maximizing the combined likelihood of the configurational data and the sticky N-3 mutant accessibility fold-changes (see Materials and methods).

Figure 2—figure supplement 6. Stage 2: logarithmic likelihood ratio histograms.

Histograms of the logarithmic likelihood ratio with respect to the perfect fit likelihood in stage 2, ${\displaystyle R_2}$, that is, using the configurational data of Brown et al., 2013 and the sticky N-3 accessibility data (Table 1). (A-D) For all models and three model subsets. 0 on the x-axis corresponds to a perfect fit. Dashed green line: value of the best model (within the subset). Red line: threshold for a ‘good’ fit of $R_{max}=6$.

Figure 2—figure supplement 7. Stage 2: prefactor histograms.

Histograms of the four prefactor values for the reaction sets a3, s23, d3 and s32 for both sticky mutants and all 15 models with maximum likelihood above the threshold in stage 2.

Figure 2—figure supplement 8. Stage 2: models with likelihood above the threshold.

Same as in Figure 2—figure supplement 4, but showing the regulated on-off-slide models with likelihood above the threshold with up to seven parameters after stage 2, with the stage 2 log10 likelihood ratios, ${\displaystyle R_2}$, written below the model numbers.